It is common for large designs to have heterogeneous components interacting with each other. These components often follow a particular model of computation such as controllers modeled using state machines, signal processing filters modeled as data flow and event-based components using discrete-event. Hence, there are several academic and industrial attempts at incorporating heterogeneity into the design flow, primarily in system level design languages and frameworks for modeling and simulation. A variety of attempts are proposed such as extending simulation kernels for existing frameworks and simply using language constructs to mimic other models of computation. However, the benefit of one over the other is not apparent to the designer and thus not clear which of the two is a better strategy for EDA tools to integrate. In this paper we argue whether Deep heterogeneity (kernel-level) or Shallow heterogeneity (language-level) is a suitable strategy for introducing heterogeneity in system level design languages and frameworks.