We present the Spin Hall Effect (SHE) Computational Random Access Memory (CRAM) for in-memory computation, incorporating considerations at the device, gate, and functional levels. For two specific applications (2-D convolution and neuromorphic digit recognition), we show that SHE-CRAM is 3× faster and has over 4× lower energy than a previously proposed STT-based CRAM implementation and is over 2000× faster and at least 130× more energy-efficient than state-of-the-art near-memory processing.