The process of classifying the tokens based on their syntactic properties is known as Parts-of-Speech (POS). The present study centred on proposing a novel POS tagging model for the Tamil language, utilizing a Conditional Random Fields (CRFs)-based approach. CRFs allows the model to acquire contextual and linguistics information to improve tagging accuracy. Two different datasets are used for the analysis purpose of the study. The Universal Dependency POS tagset is used to annotate the first dataset consist of 8,600 tokens, collected from the NLP research Centre at the University of Moratuwa. The second Tamil data set is obtained from online news sources, having the same number of 8,600 tokens as in the former corpus. The obtained data is further annotated using the Stanza Parts of Speech annotator. The study utilized the same tagset for ensuring the reliable stability in evaluation of the data. During training, the model learns from the given datasets. The study also tried to find out a comparative analysis of the proposed model performance in order to estimate its accuracy and effectiveness. As evaluation metrics to assess the strength and limitation of the proposed CRFs model the researcher utilized accuracy, precision, F1-score, and recall. The analysis of the both datasets concludes the findings as follows; the first dataset got an overall performance of 90.48% accuracy, having precision score 89.07%, recall 90.48% and F1-score 89.07%. The second dataset having accuracy of 84.00%, along with precision, recall, and F1-scores values of 83.92%, 83.97%, and 83.87%, respectively. The performance of the model tends to be more reliable on common classes like DET, NOUN, and PUNCT on both datasets.