In this lecture, we will present the dataflow model of computation (MoC) as a way to solve the abstraction vs. efficiency problem in the context of high-level synthesis of stream-processing applications on reconfigurable hardware such as FPGAs. We will show in particular how purely data-driven computations can be expressed in this model, removing the need for global state storage and synchronisation. We will then introduce the CAPH language as an incarnation of these principles and illustrate how concepts and techniques drawn from high-level functional programming nicely fit and support these ideas. This will be demonstrated using several examples taken from the embedded vision application domain.