Adjiashvili introduced network design in a non-uniform fault model: the edge set of a given graph is partitioned into safe and unsafe edges. A vertex pair $(s,t)$ is $(p,q)$-flex-connected if $s$ and $t$ have $p$ edge-connectivity even after the removal of any $q$ unsafe edges. Given a graph $G$, the goal is to choose a min-cost subgraph $H$ of $G$ that has desired flex-connectivity for a given set of vertex pairs. This model generalizes the well-studied edge-connectivity based network design, however, even special cases are provably much harder to approximate. The approximability of network design in this model has been mainly studied for two settings of interest: (i) single pair setting under the names FTP and FTF (fault tolerant path and fault tolerant flow), (ii) spanning setting under the name FGC (flexible graph connectivity). There have been several positive results in these papers. However, despite similarity to the well-known network design problems, this new model has been challenging to design approximation algorithms for, especially when $p,q \ge 2$. We obtain two results that advance our understanding of algorithm design in this model. 1. We obtain a $5$-approximation for the $(2,2)$-flex-connectivity for a single pair $(s,t)$. Previously no non-trivial approximation was known for this setting. 2. We obtain $O(p)$ approximation for $(p,2)$ and $(p,3)$-FGC for any $p \ge 1$, and for $(p,4)$-FGC for any even $p$. We obtain an $O(q)$-approximation for $(2,q)$-FGC for any $q \ge 1$. Previously only a $O(q \log n)$-approximation was known for these settings. Our results are obtained via the augmentation framework where we identify a structured way to use the well-known $2$-approximation for covering uncrossable families of cuts. Our analysis also proves corresponding integrality gap bounds on an LP relaxation that we formulate.