In this thesis we present nn-X, a scalable custom hardware architecture that is capable of processing convolutional neural networks in real time. nn-X is a low-powered mobile system for accelerating convolutional neural networks. The nn-X system comprises a host processor, a coprocessor and memory. The nn-X coprocessor efficiently implements pipelined operators and exploits a large amount of parallelism to deliver very high performance per unit power consumed. The prototyping platform used in this work consumes 8 W of power for the entire platform and only 3 W for the nn-X system and memory.