I would think of tiled vector map data as multi-resolution dataset, in several means more similar to multi-resolution raster datasets than vector datasets.
But this would be interesting challenge. I assume you want to take some vector data (e.g. Shapefiles) and convert them to vector tiles. With shapefiles (and more or less any other GIS/OGR vector format what I know) you have just one resolution. You could take this and convert it to e.g. mapbox vector tile format, with given resolution (zoom level). Now to make it really usable in interactive web or mobile map you would need also lower level tiles. With ordinary rasters the lower level is made using quite clear downsampling. With vector tiles there are a lot of design-specific decisions - what layers to keep, what to remove (e.g. in usual map you have buildings in high resolution, but not in lower ones), what layers to transform or simplify. This depends on map design; so multi-zoom vector tileset is actually design-specific, not universial vector data what you can redesign anyway as you may think. So tools which create vector tiles have this kind of style (zoom/layer) definitions for conversion. You can use e.g. Mapnik to render vector data to vector tiles, it is more similar to creating raster maps than to convert vectors, I guess Mapbox works like that.
Considering these specifics adding OGR driver to write (and read) e.g. mapbox vector tiles, or mapsforge format should be doable. Note also that there is no single standard tiled vector map, there are some known vendor-specific formats, most popular ones seem to be mapbox vector tiles and mapsforge. If anyone is already working on this, then I would be interested also.
> Thank you Jukka for your interest.
> In my system, I intend to do that, specifically for our huge road map.
Unfortunately this is rather a light-weight specification. Read what Jaak
Laineste wrote, study what others have done, and think about your own use
case. Would it be read-only or will users send back updates? What to do with
long roads at tile boundaries - split, put the whole road to the tile which
has most of it, or include the road in all intersecting tiles? Are you
delivering just original data in full resolution or do you plan to deliver
simplified versions to be used as overviews?
In the most simple case you can cut tiles on-the-fly or pre-processed with
ogr2ogr http://www.gdal.org/ogr2ogr.html. Read about -split and various clip