This paper describes the design and implementation of a convolutional neural network for 26 handwritten letters recognition on a regular embedded computer. Recognition task is carried out using a customized convolutional neural network, designed to work with low computational resources. Furthermore, training was conducted on the recently published dataset EMNIST. The experimental results show that the proposed neural network achieves an outstanding accuracy rate compared to similar architectures, also, inference shows a fast response time on a Raspberry Pi 3 board.