In one embodiment, a subset of a set of streaming kernels of an application is selected for implementation on a reconfigurable processor. The streaming kernels are selected by first forming a stream flow graph of the application by parsing a program of instructions of the application, the stream flow graph having kernel nodes and edges, and determining benefit and cost values for each kernel node in the stream flow graph. Next, a subset of the kernel nodes that maximizes a weighted sum of the benefits values is selected, subject to a constraint that the sum of cost values is not greater than a prescribed value for the reconfigurable processor.