Recommender systems have become popular in many business systems, including the entertainment and music industries. Music apps like Spotify, YouTube Music, Amazon Music, and Apple Music have embedded recommendation engines to enhance user base and customer experience. Kannada music, largely known to Indians and a bit less known to the international community, has a wide musical audience. It has a rich musical heritage with many great singers and musicians contributing to its vast musical repertoire spanning various genres. Current recommendation systems in most music apps are based on metadata classification using album, movie, singer, year, etc. They lack abilities to finer genre classification for soft Kannada music types like ‘Janapada’, ‘Bhavageethe’, and ‘Classical’, missing to attract a certain segment of the musical audience. According to the study survey, there hasn’t been any noteworthy work done on the genres of soft music, nor has there been any use of contemporary methods for classifying genres or the availability of datasets for experimenting. The proposed work is aimed at the Music Genre classification of Kannada soft music. The methodology involves the creation of a Kannada music audio dataset, application of audio analytics for feature extraction, feature analysis, and training of Machine Learning (ML) and Deep Learning (DL) models. Techniques of hyperparameter tuning, cross-validation, feature importance analysis, and Principal Component Analysis (PCA) have been applied to optimize the trained models. Through this work, a Kannada soft music dataset of 1500 audio clips is created. Various ML and DL models are trained, and the performance of models is compared. and a classification accuracy of 96.05% is achieved. Proposed applications include a music recommendation system and Teacher-in machines.