Development of new products commonly relies on existing designs where hardware and software are modeled separately and their interactions investigated late at verification time. Composing hardware with software yields more than the sum of the two parts and solid design dedicates a great deal of time and money to understand their interactions. This paper presents a novel modeling methodology based on DiplodocusDF, a UML Model-Driven Engineering approach for the design of heterogeneous data processing systems. Our methodology separately models conflicting aspects such as communications versus computations, dataflow versus controlflow and allows for their impact on the overall system's performance to be separately analyzed at various abstraction levels. Our methodology is applied to a case study which shows how it is possible to study interactions parallel to design, thus greatly reducing time-to-market of new products and comprehension of existing designs.