The deployment of neural networks in resource-constrained devices is not a straightforward task. It requires employing lighter networks while trying to maintain accuracy, which can be achieved through model acceleration and compression techniques such as pruning, quantization or tensor decomposition. Selecting the most suitable method is non-trivial and heavily depends on the requirements of the problem at hand. Each technique alters different aspects of the original model, potentially impacting the performance or even the convergence of the model, so it is key to characterize the problem to select the most convenient technique. This paper presents MAtCHelper, an open-source visual tool to select the most suitable technique based on the problem characterisation in which is applied. The tool allows to characterise the problem through key performance indicators and assists in the selection of techniques. MAtCHelper combines multiple acceleration and compression techniques with orthogonal behaviours, exploiting the compatibility and synergies among methods. Overall, MAtCHelper allows users to reduce the complexity of deep neural networks and execute them on resource-constrained devices, based on the problem needs.