In this paper, we present a class of cascaded nonlinear models for complex-valued system identification, aimed at baseband modeling of nonlinear radio systems. The proposed models consist of serially connected elementary linear and nonlinear blocks, with the nonlinear blocks implemented as uniform spline-interpolated look-up tables (LUT) and the linear blocks as FIR filters. Wiener, Hammerstein, and Wiener-Hammerstein models are built, and simple but efficient gradient based adaptive learning rules are derived for all the models. This approach leads to remarkably simple solutions in terms of computational complexity, making the techniques suitable for realtime implementation. The proposed methods are then applied to full-duplex self-interference cancellation and digital predistortion in various real-life scenarios. First, evaluations with measured data from an in-band full-duplex prototype working at 2.4 GHz ISM band show that the algorithms are capable of obtaining similar cancellation performance as existing state-of-theart solutions, regardless of the clearly reduced complexity. Second, a mmW active antenna array working at 28 GHz center frequency is digitally predistorted with the proposed solutions. The unwanted emissions and nonlinear distortion are suppressed to similar levels as with other state-of-the art solutions, and the corresponding linearity specifications are fulfilled in all cases, while the processing complexity is again drastically reduced.