Deep Learning is a field of Artificial Intelligence that has recently become very important when building intelligent systems capable of classifying images automatically and with very high accuracy. The goal is always to build a neural network architecture with high accuracy which is especially important when used for medical applications, such as skin cancer detection. However, as new Deep Learning architectures are being created these new models become more complex, increasing their depth and size due to the number of hyper-parameters involved. This increase becomes a problem when trying to deploy them in hardware with limited capacity in terms of storage and memory, such as Edge devices. In this article we create a new convolutional neural network architecture that has an accuracy of 99% for skin cancer classification, and we then compress it using a technique called Magnitude-based weight pruning to reduce its size to 65% while retaining the same accuracy.