diff --git a/src/ds/thread_messaging.h b/src/ds/thread_messaging.h
index 9f404c9e8e3c..84983cb18e10 100644
--- a/src/ds/thread_messaging.h
+++ b/src/ds/thread_messaging.h
@@ -18,50 +18,32 @@ extern std::map<std::thread::id, uint16_t> thread_ids;
 
 namespace enclave
 {
-  struct ThreadMsg
+  const uint64_t magic_const = 0xba5eball;
+  struct alignas(8) ThreadMsg
   {
     void (*cb)(std::unique_ptr<ThreadMsg>);
     std::atomic<ThreadMsg*> next = nullptr;
-    uint64_t padding[14];
+    uint64_t magic = magic_const;
+
+    ThreadMsg(void (*_cb)(std::unique_ptr<ThreadMsg>)) : cb(_cb) {}
+
+    virtual ~ThreadMsg()
+    {
+      assert(magic == magic_const);
+    }
   };
 
   template <typename Payload>
-  struct Tmsg
+  struct alignas(8) Tmsg : public ThreadMsg
   {
+    Payload data;
+
     Tmsg(void (*_cb)(std::unique_ptr<Tmsg<Payload>>)) :
-      cb(reinterpret_cast<void (*)(std::unique_ptr<ThreadMsg>)>(_cb)),
-      next(nullptr)
-    {
-      check_invariants();
-    }
+      ThreadMsg(reinterpret_cast<void (*)(std::unique_ptr<ThreadMsg>)>(_cb))
 
-    void (*cb)(std::unique_ptr<ThreadMsg>);
-    std::atomic<ThreadMsg*> next;
-    union
-    {
-      Payload data;
-      uint64_t padding[14];
-    };
+    {}
 
-    static void check_invariants()
-    {
-      static_assert(
-        sizeof(ThreadMsg) == sizeof(Tmsg<Payload>), "message is too large");
-      static_assert(
-        sizeof(Payload) <= sizeof(ThreadMsg::padding),
-        "message payload is too large");
-      static_assert(std::is_pod<Payload>::value, "data should be a pod");
-
-      static_assert(
-        offsetof(Tmsg, cb) == offsetof(ThreadMsg, cb),
-        "Expected cb at start of struct");
-      static_assert(
-        offsetof(Tmsg, next) == offsetof(ThreadMsg, next),
-        "Expected next after cb in struct");
-      static_assert(
-        offsetof(Tmsg, data) == offsetof(ThreadMsg, padding),
-        "Expected payload after next in struct");
-    }
+    virtual ~Tmsg() = default;
   };
 
   static void init_cb(std::unique_ptr<ThreadMsg> m)
@@ -168,6 +150,7 @@ namespace enclave
   public:
     static ThreadMessaging thread_messaging;
     static std::atomic<uint16_t> thread_count;
+    static const uint16_t main_thread = 0;
 
     static const uint16_t max_num_threads = 64;
 
diff --git a/src/enclave/endpoint.h b/src/enclave/endpoint.h
index 0b5140d2c54c..4c630e56c9c0 100644
--- a/src/enclave/endpoint.h
+++ b/src/enclave/endpoint.h
@@ -6,7 +6,7 @@
 
 namespace enclave
 {
-  class Endpoint
+  class Endpoint : public std::enable_shared_from_this<Endpoint>
   {
   public:
     virtual ~Endpoint() {}
diff --git a/src/enclave/httpendpoint.h b/src/enclave/httpendpoint.h
index 162c38d8837f..00f420757a5c 100644
--- a/src/enclave/httpendpoint.h
+++ b/src/enclave/httpendpoint.h
@@ -27,7 +27,23 @@ namespace enclave
       p(parser_type, *this)
     {}
 
+    static void recv_cb(std::unique_ptr<enclave::Tmsg<SendRecvMsg>> msg)
+    {
+      reinterpret_cast<HTTPEndpoint*>(msg->data.self.get())
+        ->recv_(msg->data.data.data(), msg->data.data.size());
+    }
+
     void recv(const uint8_t* data, size_t size) override
+    {
+      auto msg = std::make_unique<enclave::Tmsg<SendRecvMsg>>(&recv_cb);
+      msg->data.self = this->shared_from_this();
+      msg->data.data.assign(data, data + size);
+
+      enclave::ThreadMessaging::thread_messaging.add_task<SendRecvMsg>(
+        execution_thread, std::move(msg));
+    }
+
+    void recv_(const uint8_t* data, size_t size)
     {
       recv_buffered(data, size);
 
@@ -95,7 +111,23 @@ namespace enclave
       session_id(session_id)
     {}
 
+    static void send_cb(std::unique_ptr<enclave::Tmsg<SendRecvMsg>> msg)
+    {
+      reinterpret_cast<HTTPServerEndpoint*>(msg->data.self.get())
+        ->send_(msg->data.data);
+    }
+
     void send(const std::vector<uint8_t>& data) override
+    {
+      auto msg = std::make_unique<enclave::Tmsg<SendRecvMsg>>(&send_cb);
+      msg->data.self = this->shared_from_this();
+      msg->data.data = data;
+
+      enclave::ThreadMessaging::thread_messaging.add_task<SendRecvMsg>(
+        execution_thread, std::move(msg));
+    }
+
+    void send_(const std::vector<uint8_t>& data)
     {
       // This should be called with raw body of response - we will wrap it with
       // header then transmit
@@ -133,6 +165,76 @@ namespace enclave
       flush();
     }
 
+    struct HandleProcessCbMsg
+    {
+      std::shared_ptr<Endpoint> self;
+      std::shared_ptr<JsonRpcContext> rpc_ctx;
+      std::shared_ptr<RpcHandler> frontend;
+    };
+
+    static void handle_process(
+      std::unique_ptr<enclave::Tmsg<HandleProcessCbMsg>> msg)
+    {
+      reinterpret_cast<HTTPServerEndpoint*>(msg->data.self.get())
+        ->handle_message_main_thread(msg->data.rpc_ctx, msg->data.frontend);
+    }
+
+    struct SendResponseVectCbMsg
+    {
+      std::vector<uint8_t> d = {};
+      http_status status;
+      std::shared_ptr<Endpoint> self;
+    };
+
+    static void send_response_vect(
+      std::unique_ptr<enclave::Tmsg<SendResponseVectCbMsg>> msg)
+    {
+      reinterpret_cast<HTTPServerEndpoint*>(msg->data.self.get())
+        ->send_response(msg->data.d);
+    }
+
+    void handle_message_main_thread(
+      std::shared_ptr<JsonRpcContext>& rpc_ctx,
+      std::shared_ptr<RpcHandler>& search)
+    {
+      try
+      {
+        auto response = search->process(rpc_ctx);
+
+        if (!response.has_value())
+        {
+          // If the RPC is pending, hold the connection.
+          LOG_TRACE_FMT("Pending");
+          return;
+        }
+        else
+        {
+          // Otherwise, reply to the client synchronously.
+          LOG_TRACE_FMT("Responding");
+          auto msg = std::make_unique<enclave::Tmsg<SendResponseVectCbMsg>>(
+            &send_response_vect);
+          msg->data.status = HTTP_STATUS_OK;
+          msg->data.self = this->shared_from_this();
+          msg->data.d = response.value();
+          enclave::ThreadMessaging::thread_messaging
+            .add_task<SendResponseVectCbMsg>(execution_thread, std::move(msg));
+        }
+      }
+      catch (const std::exception& e)
+      {
+        auto msg = std::make_unique<enclave::Tmsg<SendResponseVectCbMsg>>(
+          &send_response_vect);
+        msg->data.status = HTTP_STATUS_INTERNAL_SERVER_ERROR;
+        msg->data.self = this->shared_from_this();
+
+        std::string err_msg = fmt::format("Exception:\n{}\n", e.what());
+        msg->data.d.assign(err_msg.begin(), err_msg.end());
+
+        enclave::ThreadMessaging::thread_messaging
+          .add_task<SendResponseVectCbMsg>(execution_thread, std::move(msg));
+      }
+    }
+
     void handle_message(
       http_method verb,
       const std::string& path,
@@ -246,20 +348,13 @@ namespace enclave
         rpc_ctx->method = method_s;
         rpc_ctx->actor = actor;
 
-        auto response = search.value()->process(rpc_ctx);
-
-        if (!response.has_value())
-        {
-          // If the RPC is pending, hold the connection.
-          LOG_TRACE_FMT("Pending");
-          return;
-        }
-        else
-        {
-          // Otherwise, reply to the client synchronously.
-          LOG_TRACE_FMT("Responding");
-          send_response(response.value());
-        }
+        auto msg =
+          std::make_unique<enclave::Tmsg<HandleProcessCbMsg>>(&handle_process);
+        msg->data.self = this->shared_from_this();
+        msg->data.rpc_ctx = rpc_ctx;
+        msg->data.frontend = search.value();
+        enclave::ThreadMessaging::thread_messaging.add_task<HandleProcessCbMsg>(
+          enclave::ThreadMessaging::main_thread, std::move(msg));
       }
       catch (const std::exception& e)
       {
diff --git a/src/enclave/tlsendpoint.h b/src/enclave/tlsendpoint.h
index fdc5f4997ad9..a8c0b37d17aa 100644
--- a/src/enclave/tlsendpoint.h
+++ b/src/enclave/tlsendpoint.h
@@ -9,6 +9,8 @@
 #include "tls/context.h"
 #include "tls/msg_types.h"
 
+#include <exception>
+
 namespace enclave
 {
   class TLSEndpoint : public Endpoint
@@ -16,6 +18,7 @@ namespace enclave
   protected:
     ringbuffer::WriterPtr to_host;
     size_t session_id;
+    size_t execution_thread;
 
     enum Status
     {
@@ -62,6 +65,15 @@ namespace enclave
       ctx(move(ctx_)),
       status(handshake)
     {
+      if (enclave::ThreadMessaging::thread_count > 1)
+      {
+        execution_thread =
+          (session_id_ % (enclave::ThreadMessaging::thread_count - 1)) + 1;
+      }
+      else
+      {
+        execution_thread = 0;
+      }
       ctx->set_bio(this, send_callback, recv_callback, dbg_callback);
     }
 
@@ -195,12 +207,42 @@ namespace enclave
 
     void recv_buffered(const uint8_t* data, size_t size)
     {
+      if (thread_ids[std::this_thread::get_id()] != execution_thread)
+      {
+        throw std::exception();
+      }
       pending_read.insert(pending_read.end(), data, data + size);
       do_handshake();
     }
 
+    struct SendRecvMsg
+    {
+      std::vector<uint8_t> data;
+      std::shared_ptr<Endpoint> self;
+    };
+
+    static void send_raw_cb(std::unique_ptr<enclave::Tmsg<SendRecvMsg>> msg)
+    {
+      reinterpret_cast<TLSEndpoint*>(msg->data.self.get())
+        ->send_raw_thread(msg->data.data);
+    }
+
     void send_raw(const std::vector<uint8_t>& data)
     {
+      auto msg = std::make_unique<enclave::Tmsg<SendRecvMsg>>(&send_raw_cb);
+      msg->data.self = this->shared_from_this();
+      msg->data.data = data;
+
+      enclave::ThreadMessaging::thread_messaging.add_task<SendRecvMsg>(
+        execution_thread, std::move(msg));
+    }
+
+    void send_raw_thread(std::vector<uint8_t>& data)
+    {
+      if (thread_ids[std::this_thread::get_id()] != execution_thread)
+      {
+        throw std::runtime_error("running from incorrect thread");
+      }
       // Writes as much of the data as possible. If the data cannot all
       // be written now, we store the remainder. We
       // will try to send pending writes again whenever write() is called.
@@ -222,11 +264,21 @@ namespace enclave
 
     void send_buffered(const std::vector<uint8_t>& data)
     {
+      if (thread_ids[std::this_thread::get_id()] != execution_thread)
+      {
+        throw std::runtime_error("running from incorrect thread");
+      }
+
       pending_write.insert(pending_write.end(), data.begin(), data.end());
     }
 
     void flush()
     {
+      if (thread_ids[std::this_thread::get_id()] != execution_thread)
+      {
+        throw std::runtime_error("running from incorrect thread");
+      }
+
       do_handshake();
 
       if (status != ready)
@@ -444,6 +496,10 @@ namespace enclave
 
     int handle_recv(uint8_t* buf, size_t len)
     {
+      if (thread_ids[std::this_thread::get_id()] != execution_thread)
+      {
+        throw std::runtime_error("running from incorrect thread");
+      }
       if (pending_read.size() > 0)
       {
         // Use the pending data vector. This is populated when the host