Parallel computer architectures based on such processors as the Texas Instruments TMS320C40, or the Analog Devices ADSP21060, are characterized by high performance and I/O bandwidth, flexible topology and low cost making them ideal for embedded parallel signal processing and instrumentation applications. However, the software of such systems is difficult to manage by conventional software engineering methods because of the complexity of the large-scale parallel application and the flexibility of the hardware topology. This paper discusses the adoption of a model-integrated programming environment, the Multigraph Architecture (MGA), to this domain. By using the MGA, the parallel application is automatically synthesized from high-level system models and assigned to the available network of processors.