Homomorphic Encryption (HE) makes it possible to compute on encrypted data without decrypting, thereby enabling secure data analysis. The goal of this research is to evaluate the use of HE schemes like Paillier, BFV and CKKS in sensitive real-world scenarios such as health care and financial industries. The study focuses on the performance, scalability and security implications of applying these encryption methods to encrypted machine learning models. Performance is evaluated using various criteria, including encryption time, computation overhead, model accuracy and memory utilization. A hybrid approach using two or more schemes is suggested to strike a balance between security and computational efficiency. Paillier shows a lower memory footprint, whereas CKKS delivers a stronger model accuracy for approximate calculations. The hybrid approach outperforms each of its components by offering improved processing speed and better model accuracy. Security evaluations also demonstrate that confidentiality is preserved against potential ciphertext attacks during computation. The findings in this study can be valuable for implementing HE in sensitive contexts, facilitating the development of secure data analysis techniques. These outcomes demonstrate how HE can be effectively used to ensure secure and privacy-preserving calculations in a variety of practical applications.