When CNNs are deployed on edge devices, often a huge number of dedicated hardware multiply-accumulate (MAC) units are not available to process massive MAC operations in CNNs. Instead, CPUs exist in nearly all these devices. CPUs themselves are not good at executing such mathematical operations on a large scale, since they opt more to execute control flow logic. To execute CNNs on CPUs eGiciently, it is critical to convert their MAC operations into logic flows. In this master thesis, the execution of a convolutional neural network (CNN) will be converted to logic flows, so that it can be executed with low latency and low energy on CPUs.