The accurate identification of script styles in historical manuscripts is crucial for gaining insights into their historical context and significance. This study proposes an integrated system that combines manual and machine learning features to effectively identify script styles within manuscripts, utilizing the ClAMM dataset. The system comprises three primary steps: preprocessing the dataset using denoising and binarization techniques, extracting manual features using the Harris detector, and performing script classification using pretrained CNN models. By merging manual feature engineering with advanced deep learning techniques, our system showcases its ability to accurately recognize script styles competing with state-of-the-art methods with accuracy of 89.2%. These results not only validate the effectiveness of our approach but also contribute significantly to the broader advancement of script classification in historical manuscript analysis.