Object detection, identification and classification techniques have seen many variants and improvements over past two decades. Together with Internet of Things (IoT) devices, improved computational algorithms and cloud support, real-time classification with low-cost devices has already been achieved. This paper discusses the real-time object detection and classification using Microsoft Custom Vision multi-class Machine Learning (ML) model operating at the Edge of IoT network. This paper further examines the use of virtual dockers or containers at the IoT edge devices for better security and isolation by decoupling physical hardware as well that supports multiple applications and services on a single hardware. The experiments are performed using emulated and simulated IoT devices on Microsoft Azure IoT platform for real-time object classification using Custom Vision Machine Learning (ML) models run directly from the edge device. The experimental results are further discussed to validate the model accuracy and its implementation in a future Smart City surveillance environment.