Homomorphic encryption (HE) allows computations on encrypted data, safeguarding sensitive information and enabling secure data processing in untrusted cloud environments. However, the computational complexity of current HE schemes has hindered its practical deployment. While CPUs and GPUs cannot efficiently handle HE’s integer-based parallelisable workloads, FPGAs offer a compelling solution, offering both flexibility and computational efficiency. In this paper, we present an FPGA-based implementation of homomorphic 2D convolution, targeting the BFV scheme. By focusing on optimising ciphertext-plaintext multiplications and ciphertext-ciphertext additions, we reduce computational overhead while maintaining data security. Our work applies plaintext convolutional filters to encrypted images, forming the foundation for feature extraction nodes in deep learning models. This approach enables secure processing of encrypted data, making it suitable for privacy-preserving machine learning applications.