We propose a methodology, and its embodiment into a design flow, to realize execution platforms for high-performance building applications. This is an example of a class of cyberphysical systems where a network of sensors, controllers, and actuators must be designed under physical spatial constraints to implement various types of signal processing and control tasks. In our approach, the applications are specified using the dataflow model of computation while the building dictates the physical constraints, including the position of sensors and actuators. We present a rigorous formulation of the design-space exploration problem and we propose to solve it by progressing through a sequence of refinement steps from specification to detailed implementation. Two key steps are the synthesis of the computation platform and the synthesis of the communication network. Combined, they allow us to automatically derive an optimal implementation through the selection and composition of processing and networking elements from given technology libraries. We demonstrate the applicability of our approach by comparing it to the manual design of a given case study: the real-time estimation of building occupancy using a network of video cameras.