Dynamic languages such as Python, Ruby, or R rely heavily on native C/C++ functions using a foreign function interface. While they significantly speed up baseline performance, these functions are opaque to a Just-in-time compiler — parameter and return types, side effects, and argument counts are not exposed for optimization. We propose an approach that analyzes native code in the context of the host VM, opening up the semantics of native code without the need for manual transformation. From this we identify three use cases: a) inferring precise types and side effects for JIT specialization, b) modeling the dynamic language semantics in the same IR, enabling whole-program analysis, and c) extracting specialized native code slices that only capture the main logic for subsets of inputs.