Email has become the de facto standard for communication and attackers are taking this opportunity to send junk emails, making it difficult for recipients to distinguish between spam and not spam. In this article, our goal is to build a spam filtering framework by considering the best-known approaches. Pretty Good Privacy (PGP) is a well-known cryptography technique for secure communication via encrypted and decrypted emails. To make communication more secure, Homomorphic Encryption (HE) is used for processing emails without revealing their contents. Email Spam Filtering is another approach for classifying whether emails are spam or not spam. For deciding the suitable classifier from Support Vector Machine, Random Forest, Logistic Regression, and KNN, we consider three cases where the input data consists of the ciphertext of the email as well as the metadata of the email content. Case 1 simply performs the classification over the plaintext of the email. Case 2 performs the classification over the encrypted email without using any metadata while Case 3 considers the encrypted email as well as the metadata of the email content to perform the classification. KNN gives the best accuracy of 92.393% in plain body but it fails to give the best accuracy in the case of encrypted email. In all the cases of encrypted email, Logistic Regression gives the best accuracy compared to other classifiers with an accuracy of 83.219% in Case 2, 83.515% accuracy in Case 3.1, and 83.988% accuracy in Case 3.2, and 86.061% in Case 3.3.