A system having an off-premises proxy server residing in a cloud computing environment and backend servers residing in an enterprise computing environment are provided. Requests received by the off-premises proxy server for access to a first, non-publicly accessible backend server are routed to a tunnel server which stores the request and waits to be polled by a tunnel agent connected to the first backend server. When the tunnel server is polled, the request is forwarded through an HTTP tunnel to the tunnel agent, which forwards it to the backend server for processing. Responsive information is returned to the tunnel agent, which forwards it through the HTTP tunnel to the tunnel server and returned through the off-premises proxy server to the remote application. Requests for access to a first, publicly accessible backend server are routed by the off-premises proxy server directly to the backend server for processing and return of responsive information.