We present the results of a model-independent investigation of the rest-frame UV spectra from a comprehensive sample of $394$ quasars in the redshift range $1.5\leq z \leq 7.5$. We fit the main Broad Emission Lines (BELs) in the rest-frame range $1280 \text{ \AA} \leq \lambda \leq 3000 \text{ \AA}$ (OI,CII,SiIV, CIII], CIV and MgII) with a lightly-supervised spline fitting technique. Redshifts are derived from the peaks of each fitted BEL and used to compute relative velocity shifts. We show that our method gives unbiased velocity shifts and is insensitive to spectral resolution and instrumental parameters. It is found that the average blueshift of the CIV line with respect to several low-ionisation lines in luminosity-matched samples does not significantly evolve over $1.5\gtrsim z\gtrsim6$. However, the average blueshift increases significantly by a factor $\sim 2.5$ at $z\gtrsim 6$. We propose that this redshift evolution can be explained by CIV winds launched perpendicularly to an accretion disk with increased torus opacity at high-redshift, coupled with a potential orientation-driven selection bias. Our results open new exciting avenues of investigation into young quasars in the reionisation epoch.