The last decade has witnessed tremendous advances in cloud computing, IoT, and computer vision. IoT computer vision brings powerful capabilities that enables society to tackle complex problems such as autonomous driving vehicles, smart cities, public safety, and interactive healthcare. However, the field faces challenges of large data streams, complex processing, low latency requirements, and data privacy concerns. Processing at the Edge vastly reduces the data that needs to be sent to the cloud (by a factor of 1000). Additionally, Edge processing results in lowered application latency and sensitive video streams are confined to the privacy perimeter of the end-user (for example, homes, hospitals, etc.). However, current IoT system software infrastructures are designed for low data rate sensor applications and do not satisfy the demanding needs of computer vision-based IoT.In this thesis, we design and implement an Edge gateway targeted specifically at emerging IoT computer vision applications. The proposed Edge gateway, which we call VEI, enables realization of multiple vision algorithms at the Edge from a single camera stream. Furthermore, unlike existing Edge gateways, VEI is vendor-neutral, and capable of connecting to any Cloud provider. This allows for increased application resilience, lowers costs, and avoids Cloud vendor lock-in. We experimentally evaluate the performance of VEI for canonical object detection applications. Public clouds considered in this work include those from Amazon (AWS) and Google (GCP).