Memristive devices arranged in cross-bar architectures have shown great promise to facilitate the acceleration and improve the power efficiency of Deep Learning (DL) systems for deployment in near-sensor Internet-of-Things (IoT) edge devices. These cross-bar architectures can be used to implement various in-memory computing operations, which are used extensively in Deep Neural Networks (DNNs) and Convolutional Neural Networks (CNNs), such as Multiply-accumulate (MAC) and convolution operations. Currently, there is a lack of a general high-level simulation platform that can integrate behavioural or experimental memristive device models into cross-bar architectures. This paper presents such a framework, which integrates directly with the well-known PyTorch Machine Learning (ML) library. We perform simulations using it to demonstrate the degregation in performance in which non-ideal devices introduce to Memristive DNNs (MDNNs) using VGG-16 and CIFAR-10. Our open source MemTorch framework can be used by circuits and system designers to conveniently build customized large-scale simulation platforms, as an intermediary step before circuit-level realization.