The increasing complexity of distributed systems, where het- erogeneous systems are composed to form systems of systems, pose new development challenges. How can core middleware services, e.g. event communication, resource discovery, etc. be deployed and optimised in an end-to-end manner? Further, how can important properties such as inter- operability be managed? In this paper we propose OverStar a framework that generates overlay network based solutions from high-level specifica- tions in order to answer these questions. A middleware service is specified as a self-managing overlay network across heterogeneous systems; timed automata specify how the topology of the network is constructed and the data is exchanged. The key contribution is the open access to individual overlay nodes in order to specify additional flow logic, e.g. the transla- tion of messages to support end-to-end interoperability or the filtering of heterogeneous messages to optimise event dissemination. We evaluate OverStar using service discovery and event communication case studies; these demonstrate the ability to compose heterogeneous systems, achieve end-to-end interoperability and simplify the developer's task. Further, a performance evaluation highlights optimisations that can be achieved.