Dynamic analysis in IoT environments is often hindered by the lack of knowledge in certain critical components. This project aims to develop techniques to bridge the gap and enable effective dynamic analysis. Based on our research in traditional binaries, we found that important functionalities of components, such as dataflow-related semantics, can be inferred from observing inputs and outputs of a component. We will further the investigation of this technique in the domain of IoT components, using inference-based techniques to model components missing in virtual machines. With the model of IoT components, we can further drive the execution of firmware in dynamic analysis, and thus exposing more of its functionality to analysis, such as fuzzing and taint analysis.