Human breath analysis plays important role for diagnosis and management of pulmonary diseases to guarantee normal health. The critical task is to distinguish normal and abnormal lung sounds. This research work presents a scheme for breath analysis used to detect irregular patterns occurred in respiratory cycles due to respiratory diseases. After de-noising breath segments using wavelet de-noising method, intrinsic mode functions are extracted with complete ensemble empirical mode decomposition (CEEMD). Instantaneous frequency (IF) and instantaneous envelope are extracted to get robust features for classification. The study contains breath samples captured using smartphone under natural setting. The data set contains 255 breath cycles. For cycle classification, Bag-of-word was applied to group segments based features. The support vector machine (SVM) was applied on randomly partitioned data samples. Experiments resulted with performance accuracy of (75.21%±2) for asthmatic inspiratory cycles and (75.5%±3%) for complete Respiratory Sounds (RS) cycle with diagnostic odds ratio (DOR) of 20.61% and 13.S7% respectively.