Interoperability remains a challenging and growing problem within distributed systems. A range of heterogeneous network and middleware protocols which cannot interact with one another are now widely used, for example, the set of remote method invocation protocols, and the set of service discovery protocols. In environments where systems and services are composed dynamically, e.g. pervasive computing and systems-of-systems, the protocols used by two systems wishing to interact is unknown until runtime and hence interoperability cannot be guaranteed. In such situations, dynamic solutions are required to identify the differences between heterogeneous protocols and generate middleware connectors (or bridges) that will allow the systems to inter operate. In this paper, we present the Starlink middleware, a general framework into which runtime generated interoperability logic (in the form of higher level models) can be deployed to connect two heterogeneous protocols. For this, it provides: i) an abstract representation of network messages with a corresponding generic parser and composer, ii) an engine to execute coloured automata that represent the required interoperability behaviour between protocols, and iii) translation logic to describe the exchange of message content from one protocol to another. We show through case-study based evaluation that Starlink can bridge heterogeneous protocol types. Starlink is also compared against base-line protocol benchmarks to show that acceptable performance can still be achieved in spite of the high-level nature of the solution.