Hardware specialization has emerged as a promising paradigm for future microprocessors. Unfortunately, it is natural to develop and evaluate such architectures within end-to-end vertical silos spanning application, language/compiler, hardware design and evaluation tools, leaving little opportunity for cross-architecture analysis and innovation. This paper develops a novel program representation suitable for modeling heterogeneous architectures with specialized hardware, called the transformable dependence graph (TDG), which combines semantic information about program properties and low-level hardware events in a single representation. We demonstrate, using four example architectures from the literature, that the TDG is a feasible, simple, and accurate modeling technique for transparent specialization architectures, enabling cross-domain comparison and design-space exploration.