Paraphrasing in the context of low-resource languages suffers from scarcity of large paraphrase corpus, which impacts several downstream tasks where paraphrasing is essential. We propose a data augmentation framework that employs Round-Trip-Translation (RTT) to extend a monolingual corpus into a much larger paraphrase corpus. We incorporate sequence-to-sequence Neural Machine Translation (NMT) model, which integrates multiple decoding strategies at each stage of RTT to ensure that paraphrases are lexically diverse. In addition to this, we publicly release a T5-based Kannada paraphrasing model KnParaphraser fine-tuned on the augmented data. A 100 k sentence-level paraphrase corpus KanPara generated using our model is also released. We achieve an average BLEU score of 0.1967 and a sentence similarity score of 0.8133, evaluating on 1 k sentences. The proposed framework addresses the challenges of paraphrase generation in low-resource languages and contributes valuable resources to the field of natural language processing.