Using green foliage (or chlorophyll) to set a custom white balance deserves more consideration than I gave it in my last comment. Foliage absorbs roughly about half of the blue light and a quarter of the red light hitting the leaves. So roughly half of the blue light and 75% of the red light might be reflected back to the camera.
When a custom white balance is done with a full spectrum camera with red filter (e.g., your K610) NIR light must also be considered. Foliage reflects almost all of the NIR hitting it, so lots of NIR will come back to the camera.
The blue channel will not get much blue light because the red filter blocks it, but lots of NIR will be recorded in the blue channel. The red channel will record both the red and NIR reflected from the foliage. The red channel is often more sensitive to NIR than the blue channel is.
So the red channel will record more brightness than the blue channel which is exactly what the custom white balance procedure needs in order to exaggerate the blue channel values. So pointing the camera at foliage in daylight during the custom white balance procedure will push the results in the right direction. It probably will not push the results far enough to produce NDVI values in the correct range, but the results could be interpretable.
Instead of using foliage, using a surface that reflects lots of red, no blue, and no NIR will do more to exaggerate the blue channel values. When the custom white balance algorithm sees all that brightness in the red channel and none in the blue, it will adjust every photo by exaggerating the blue channel values.
We generally don't know how much NIR is being reflected from a colored surface, so choosing a useful surface can involve some trial and error. However, with a full spectrum camera and your K720 filter (assuming that filter passes only NIR) you can easily determine the relative NIR reflection from various surfaces. The goal should be to find a surface that is reddish but reflects no NIR, although some NIR reflection does not necessarily disqualify a surface.
This process is different from a calibration procedure which adjusts the red and blue values based on photos of targets of known reflectivity and compensation for the contamination of the red channel with NIR. Compared to calibration, this process is a clever hack which can produce potentially useful results.