Programming distributed embedded systems gives rise to a number of challenges. The distributed nodes are typically resource constrained, requiring expert systems programming skills to manage the limited power, communication bandwidth, and memory and computation constraints. The challenge of raising the level of abstraction of programming such systems without incurring too high of an execution performance penalty is thus an important one, and many approaches have been explored in the literature. The work presented in this paper investigates a framework and domain specific language (DSL) to enable programming of such systems at a global network level. The framework enables the compilation, analysis, transformation and interpretation of high-level descriptions of stream processing applications in which information is received and processed in real-time. The ongoing research aims at investigating the following directions. Firstly, a framework to allow the manipulation and multiple interpretations of a stream processor description to support the heterogeneity aspect of devices. Secondly, a language to hide the low-level detail from the developer, while allowing the developer to add hints to enable more efficient compilations. Thirdly, through transformations, different stream processing applications can be fused together to run on the same distributed network to make best of use of available resources.