Ambient intelligence and wearable computing call for innovative hardware and software technologies, including a highly capable, flexible and efficient middleware, allowing for the reuse of existing pervasive applications when developing new ones. In the considered application domain, middleware should also support self-management, interoperability among different platforms, efficient communications, and context awareness. In the on-going "everything is networked" scenario scalability appears as a very important issue, for which the peer-to-peer (P2P) paradigm emerges as an appealing solution for connecting software components in an overlay network, allowing for efficient and balanced data distribution mechanisms. In this paper, we illustrate how all these concepts can be placed into a theoretical tool, called networked autonomic machine (NAM), implemented into a NAM-based middleware, and evaluated against practical problems of pervasive computing.