The current Internet was originally designed to provide "best-effort" data transport over a wired infrastructure with end hosts utilizing a layered network stack to provide reliability, quality of service, security etc. for user applications. However, the proliferation of inelastic applications, coupled with wide spread migration towards hybrid networks utilizing wired and wireless links and the plethora of end host variants ranging from cell phones to enterprise servers necessitates the migration of more and more services away from the edges and into the network. This dissertation presents a generic and flexible framework that will enable the incremental deployment of intelligent services into the network with the aim of optimizing the end-user experience for networked applications. Our framework relies on using semantic tags embedded in the data stream that describe the type of content carried in stream and levels of service desired. The nodes in the network utilize this semantic markup to reason over the stream to identify the set of actions that need to be performed based on the policies specified. We present the results of employing our framework to various scenarios including cross layer optimization for Ultra wideband PANs, use of semantic tags to differentiate traffic into service classes and utilizing the framework to allow BGP to employ security and trust relationships in the route decision process. In addition to the development of the framework, we have also developed a network ontology that can be used to both model traffic flows and their service needs along with network topology and state. The framework employs multi-level reasoning that utilizes the meta-data carried in the streams coupled with external contextual information to determine the set of actions that need to be performed to the datastream in-transit. This reasoning is done using network policies specified by network administrators using our rule language that is easy to use, human readable and machine understandable, non-network protocol specific, and can be formally verified for consistency.