diff --git a/include/boost/thread/pthread/thread_data.hpp b/include/boost/thread/pthread/thread_data.hpp
index 46f111d9..588e997f 100644
--- a/include/boost/thread/pthread/thread_data.hpp
+++ b/include/boost/thread/pthread/thread_data.hpp
@@ -94,12 +94,15 @@ namespace boost
         struct thread_exit_callback_node;
         struct tss_data_node
         {
-            boost::shared_ptr<boost::detail::tss_cleanup_function> func;
+            typedef void(*cleanup_func_t)(void*);
+            typedef void(*cleanup_caller_t)(cleanup_func_t, void*);
+
+            cleanup_caller_t caller;
+            cleanup_func_t func;
             void* value;
 
-            tss_data_node(boost::shared_ptr<boost::detail::tss_cleanup_function> func_,
-                          void* value_):
-                func(func_),value(value_)
+            tss_data_node(cleanup_caller_t caller_,cleanup_func_t func_,void* value_):
+                caller(caller_),func(func_),value(value_)
             {}
         };
 
diff --git a/include/boost/thread/tss.hpp b/include/boost/thread/tss.hpp
index d798bef5..bcfbdf0f 100644
--- a/include/boost/thread/tss.hpp
+++ b/include/boost/thread/tss.hpp
@@ -6,8 +6,8 @@
 // (C) Copyright 2007-8 Anthony Williams
 
 #include <boost/thread/detail/config.hpp>
-#include <boost/shared_ptr.hpp>
-#include <boost/thread/detail/thread_heap_alloc.hpp>
+
+#include <boost/type_traits/add_reference.hpp>
 
 #include <boost/config/abi_prefix.hpp>
 
@@ -15,15 +15,13 @@ namespace boost
 {
     namespace detail
     {
-        struct tss_cleanup_function
+        namespace thread
         {
-            virtual ~tss_cleanup_function()
-            {}
+            typedef void(*cleanup_func_t)(void*);
+            typedef void(*cleanup_caller_t)(cleanup_func_t, void*);
+        }
 
-            virtual void operator()(void* data)=0;
-        };
-
-        BOOST_THREAD_DECL void set_tss_data(void const* key,boost::shared_ptr<tss_cleanup_function> func,void* tss_data,bool cleanup_existing);
+        BOOST_THREAD_DECL void set_tss_data(void const* key,detail::thread::cleanup_caller_t caller,detail::thread::cleanup_func_t func,void* tss_data,bool cleanup_existing);
         BOOST_THREAD_DECL void* get_tss_data(void const* key);
     }
 
@@ -34,49 +32,33 @@ namespace boost
         thread_specific_ptr(thread_specific_ptr&);
         thread_specific_ptr& operator=(thread_specific_ptr&);
 
-        struct delete_data:
-            detail::tss_cleanup_function
+        typedef void(*original_cleanup_func_t)(T*);
+
+        static void default_deleter(T* data)
         {
-            void operator()(void* data)
-            {
-                delete static_cast<T*>(data);
-            }
-        };
+            delete data;
+        }
 
-        struct run_custom_cleanup_function:
-            detail::tss_cleanup_function
+        static void cleanup_caller(detail::thread::cleanup_func_t cleanup_function,void* data)
         {
-            void (*cleanup_function)(T*);
-
-            explicit run_custom_cleanup_function(void (*cleanup_function_)(T*)):
-                cleanup_function(cleanup_function_)
-            {}
-
-            void operator()(void* data)
-            {
-                cleanup_function(static_cast<T*>(data));
-            }
-        };
+            reinterpret_cast<original_cleanup_func_t>(cleanup_function)(static_cast<T*>(data));
+        }
 
 
-        boost::shared_ptr<detail::tss_cleanup_function> cleanup;
+        detail::thread::cleanup_func_t cleanup;
 
     public:
         typedef T element_type;
 
         thread_specific_ptr():
-            cleanup(detail::heap_new<delete_data>(),detail::do_heap_delete<delete_data>())
+            cleanup(reinterpret_cast<detail::thread::cleanup_func_t>(&default_deleter))
         {}
         explicit thread_specific_ptr(void (*func_)(T*))
-        {
-            if(func_)
-            {
-                cleanup.reset(detail::heap_new<run_custom_cleanup_function>(func_),detail::do_heap_delete<run_custom_cleanup_function>());
-            }
-        }
+          : cleanup(reinterpret_cast<detail::thread::cleanup_func_t>(func_))
+        {}
         ~thread_specific_ptr()
         {
-            detail::set_tss_data(this,boost::shared_ptr<detail::tss_cleanup_function>(),0,true);
+            detail::set_tss_data(this,0,0,0,true);
         }
 
         T* get() const
@@ -87,14 +69,14 @@ namespace boost
         {
             return get();
         }
-        typename boost::detail::sp_dereference< T >::type operator*() const
+        typename add_reference<T>::type operator*() const
         {
             return *get();
         }
         T* release()
         {
             T* const temp=get();
-            detail::set_tss_data(this,boost::shared_ptr<detail::tss_cleanup_function>(),0,false);
+            detail::set_tss_data(this,0,0,0,false);
             return temp;
         }
         void reset(T* new_value=0)
@@ -102,7 +84,7 @@ namespace boost
             T* const current_value=get();
             if(current_value!=new_value)
             {
-                detail::set_tss_data(this,cleanup,new_value,true);
+                detail::set_tss_data(this,&cleanup_caller,cleanup,new_value,true);
             }
         }
     };
diff --git a/include/boost/thread/win32/thread_data.hpp b/include/boost/thread/win32/thread_data.hpp
index 2f288202..b1a550af 100644
--- a/include/boost/thread/win32/thread_data.hpp
+++ b/include/boost/thread/win32/thread_data.hpp
@@ -80,12 +80,15 @@ namespace boost
         struct thread_exit_callback_node;
         struct tss_data_node
         {
-            boost::shared_ptr<boost::detail::tss_cleanup_function> func;
+            typedef void(*cleanup_func_t)(void*);
+            typedef void(*cleanup_caller_t)(cleanup_func_t, void*);
+
+            cleanup_caller_t caller;
+            cleanup_func_t func;
             void* value;
 
-            tss_data_node(boost::shared_ptr<boost::detail::tss_cleanup_function> func_,
-                          void* value_):
-                func(func_),value(value_)
+            tss_data_node(cleanup_caller_t caller_,cleanup_func_t func_,void* value_):
+                caller(caller_),func(func_),value(value_)
             {}
         };
 
diff --git a/src/pthread/thread.cpp b/src/pthread/thread.cpp
index 1b76570b..f318635d 100644
--- a/src/pthread/thread.cpp
+++ b/src/pthread/thread.cpp
@@ -111,7 +111,7 @@ namespace boost
                                     = thread_info->tss_data.begin();
                                 if(current->second.func && (current->second.value!=0))
                                 {
-                                    (*current->second.func)(current->second.value);
+                                    (*current->second.caller)(current->second.func,current->second.value);
                                 }
                                 thread_info->tss_data.erase(current);
                             }
@@ -726,11 +726,12 @@ namespace boost
         }
 
         void add_new_tss_node(void const* key,
-                              boost::shared_ptr<tss_cleanup_function> func,
+                              detail::tss_data_node::cleanup_caller_t caller,
+                              detail::tss_data_node::cleanup_func_t func,
                               void* tss_data)
         {
             detail::thread_data_base* const current_thread_data(get_or_make_current_thread_data());
-            current_thread_data->tss_data.insert(std::make_pair(key,tss_data_node(func,tss_data)));
+            current_thread_data->tss_data.insert(std::make_pair(key,tss_data_node(caller,func,tss_data)));
         }
 
         void erase_tss_node(void const* key)
@@ -743,17 +744,19 @@ namespace boost
         }
 
         void set_tss_data(void const* key,
-                          boost::shared_ptr<tss_cleanup_function> func,
+                          detail::tss_data_node::cleanup_caller_t caller,
+                          detail::tss_data_node::cleanup_func_t func,
                           void* tss_data,bool cleanup_existing)
         {
             if(tss_data_node* const current_node=find_tss_data(key))
             {
                 if(cleanup_existing && current_node->func && (current_node->value!=0))
                 {
-                    (*current_node->func)(current_node->value);
+                    (*current_node->caller)(current_node->func,current_node->value);
                 }
                 if(func || (tss_data!=0))
                 {
+                    current_node->caller=caller;
                     current_node->func=func;
                     current_node->value=tss_data;
                 }
@@ -764,7 +767,7 @@ namespace boost
             }
             else if(func || (tss_data!=0))
             {
-                add_new_tss_node(key,func,tss_data);
+                add_new_tss_node(key,caller,func,tss_data);
             }
         }
     }
diff --git a/src/win32/thread.cpp b/src/win32/thread.cpp
index 8aceb114..e1ae050b 100644
--- a/src/win32/thread.cpp
+++ b/src/win32/thread.cpp
@@ -276,7 +276,7 @@ namespace boost
                             = current_thread_data->tss_data.begin();
                         if(current->second.func && (current->second.value!=0))
                         {
-                            (*current->second.func)(current->second.value);
+                            (*current->second.caller)(current->second.func,current->second.value);
                         }
                         current_thread_data->tss_data.erase(current);
                     }
@@ -909,11 +909,12 @@ namespace boost
         }
 
         void add_new_tss_node(void const* key,
-                              boost::shared_ptr<tss_cleanup_function> func,
+                              detail::tss_data_node::cleanup_caller_t caller,
+                              detail::tss_data_node::cleanup_func_t func,
                               void* tss_data)
         {
             detail::thread_data_base* const current_thread_data(get_or_make_current_thread_data());
-            current_thread_data->tss_data.insert(std::make_pair(key,tss_data_node(func,tss_data)));
+            current_thread_data->tss_data.insert(std::make_pair(key,tss_data_node(caller,func,tss_data)));
         }
 
         void erase_tss_node(void const* key)
@@ -923,17 +924,19 @@ namespace boost
         }
 
         void set_tss_data(void const* key,
-                          boost::shared_ptr<tss_cleanup_function> func,
+                          detail::tss_data_node::cleanup_caller_t caller,
+                          detail::tss_data_node::cleanup_func_t func,
                           void* tss_data,bool cleanup_existing)
         {
             if(tss_data_node* const current_node=find_tss_data(key))
             {
                 if(cleanup_existing && current_node->func && (current_node->value!=0))
                 {
-                    (*current_node->func)(current_node->value);
+                    (*current_node->caller)(current_node->func,current_node->value);
                 }
                 if(func || (tss_data!=0))
                 {
+                    current_node->caller=caller;
                     current_node->func=func;
                     current_node->value=tss_data;
                 }
@@ -944,7 +947,7 @@ namespace boost
             }
             else if(func || (tss_data!=0))
             {
-                add_new_tss_node(key,func,tss_data);
+                add_new_tss_node(key,caller,func,tss_data);
             }
         }
     }