In recent years, a great deal of progress has been made towards tool support for developing embedded software. Tools solve a variety of difficult problems, for example by automating error-prone implementation tasks, by eliminating redundant and inefficient constructs, and by guaranteeing the absence of certain classes of errors, such as race conditions or out-of-memory exceptions. This NSF CAREER research is about Vertically Integrated Program Analysis and Transformation (VIPAT), a new way to look at embedded software tools: as a collection of building blocks that can be connected together in different ways to support novel analyses and transformations. The existing tools become mechanisms that are controlled by a high-level policy. VIPAT is based on two main ideas. First, the vertical integration of tools that operate at various levels of abstraction, which permits high-level transformations to be precisely targeted at parts of a system whose low-level resource usage is worst. Second, a clean separation between mechanism and policy, enabling effective reuse of existing tools in new situations. This research is a step towards a world where meaningful static guarantees about program behavior can be made, and where software can be automatically specialized to meet platform- and application-specific requirements such as time and energy constraints. The high-level vision is "fearless reuse": developers should spend less time worrying about resource usage and potential failure modes of components that they reuse.