Narrowband Internet-of-Things (NB-IoT) applications can be deployed in licensed and unlicensed bands. Spectrum allocation for Narrowband IoT is often in-between licensed spectrum or available white space frequency bands and the designer is expected to meet stringent spectral mask requirements. Further, Narrowband IoT applications have range requirements in the order of few kilometers to tens of kilometers, which necessitates the use of high power amplifiers. This introduces significant non-linearity for systems which employ multiple access techniques like orthogonal frequency division multiple access (OFDMA). This effects the quality of the transmitted signal as well as introduces undesired adjacent channel emissions. In this paper, we describe a design approach for implementing a digital pre-distorter (DPD) for Narrowband IoT applications. We use a memory polynomial DPD model to trade-off performance and implementation complexity. By careful tuning of the parameters of the model, we demonstrate the system meeting the expected spectral mask and transmission quality requirements.