In the era of digital forensics, the detection of multi-modal evidence from images and videos is crucial for investigative purposes. However, privacy concerns and the need for collaborative computations among multiple parties pose significant challenges. This paper proposes a novel framework that enhances deep learning models with Homomorphic Encryption (HE) for secure multi-modal evidence detection. The system integrates Secure Multi-Party Computation (SMC) to enable forensic computations without compromising data privacy. Our approach leverages convolutional neural networks (CNNs) and recurrent neural networks (RNNs) for feature extraction from images and videos, respectively, while HE ensures that operations are performed on encrypted data. Experimental results demonstrate high accuracy in evidence detection while maintaining computational security. The framework is evaluated on benchmark datasets, showing its efficacy in real-world forensic scenarios with potential integration into existing digital investigation tools.